Accurate forecasting of residential natural gas demand is critical for utility operators managing supply logistics, pipeline scheduling, and system balancing. A new study from Qazvin Province, Iran demonstrates that incorporating infrastructure growth metrics alongside traditional meteorological data substantially improves short-term consumption predictions.
Researchers compared two forecasting datasets: one containing only weather variables, and a second augmented with a subscription growth ratio (SGR)—a measure of how rapidly new customer connections are being added to the gas network. Using five years of daily consumption data (2019–2023), they tested four machine learning models: multiple linear regression, support vector regression, Random Forest, and XGBoost.
Results show SGR ranked as the third most important predictor, behind only minimum and average temperature. When nonlinear models incorporated SGR, mean absolute error (MAE) dropped roughly 20%, while adjusted R² improved by approximately 38%. Paired statistical testing confirmed these gains were significant for support vector regression, Random Forest, and XGBoost—though surprisingly not for linear regression.
The findings suggest residential gas demand exhibits nonlinear sensitivity to infrastructure expansion. As utilities expand their networks, new customers show different consumption patterns than established ones, reflecting both behavioral differences and variations in equipment efficiency across aging versus new connections.
For gas distribution utilities, the practical implication is clear: demand forecasting models should integrate customer acquisition rates and network expansion schedules alongside weather data. This hybrid approach enables more accurate operational scheduling, better demand-supply matching, and improved system reliability. The methodology may prove transferable to other utilities managing rapid network growth in developing regions, where infrastructure expansion drives consumption patterns as strongly as seasonal temperature swings.
